Berkeley—cracking the CO2 code with massively parallel HPC

Parallel porous media adaptive mesh refinement code runs on Cray XT4 Linux supercomputer.

According to the US Energy Information Administration fossil fuel fired power plants will account for 70% of US electricity generation right out to 2035. According to George Pau of the Lawrence Berkeley Lab’s Center for Computational Sciences and Engineering (CCSE) ‘Underground carbon capture and sequestration (CCS) will be key to reducing atmospheric CO2.’ The CCSE has been studying the processes involved in CCS particularly the mixing that occurs at the CO 2/ brine interface1.

When CO2 is injected into an aquifer, it collects in a layer above the brine. Over time CO2 diffuses into the brine, causing its density to rise. This denser fluid begins to sink, generating a convective process that increases the rate of CO2 uptake. The CCSE’s Cray XT4 ‘Franklin’ supercomputer was used to figure out the time scale of the process, modeling at a much finer resolution than is usual in geological studies. New parallelized porous media adaptive mesh refinement (PMAMR) code was used for the simulation which took 2 million processor-hours running on up to 2,048 cores. Current PMAMR models are on the scale of meters, but the plan is to scale up the results to characterize a real-world scale aquifer.

LBL researcher Karsten Pruess said, ‘The CO2 from a large coal-fired power plant operating for 30 years will generate a 10 kilometer wide plume which could eventually migrate into aquifers hundreds of kilometers away. We are very interested in the long-term fate of these processes.’

Franklin comprises 9,572 compute nodes, each with a 2.3 GHz single socket quad-core AMD Opteron processor and has 8 GB memory for a total peak performance of 352 teraflops. The system deploys dual operating systems—SuSE Linux on service nodes and a lightweight ‘compute node Linux’ (CNL) on individual nodes. CNL reduces system overhead and assures scalability. A Lustre parallel file system provides some 436 TB of user disk space. The CCSE thinks that PMAMR simulations will help recover more oil from existing wells. More from

Full paper on (behind a pay wall).

This article originally appeared in Oil IT Journal 2011 Issue # 2.

For more information or to comment on this topic email here.